diff --git a/pyproject.toml b/pyproject.toml
index a750462b..744d7149 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -33,7 +33,7 @@ dependencies = [
 ]
 
 [project.scripts]
-keep-gpu = "keep_gpu.cli:main"
+keep-gpu = "keep_gpu.cli:app"
 
 [project.optional-dependencies]
 dev = [
diff --git a/src/keep_gpu/cli.py b/src/keep_gpu/cli.py
index 18e1ee54..d64e1abf 100644
--- a/src/keep_gpu/cli.py
+++ b/src/keep_gpu/cli.py
@@ -1,20 +1,78 @@
 """Console script for keep_gpu."""
 
-from .keep_gpu import run
+import os
+import time
+from typing import Optional
 
+import torch
 import typer
 from rich.console import Console
 
+from keep_gpu.global_gpu_controller.global_gpu_controller import GlobalGPUController
+from keep_gpu.utilities.logger import setup_logger
+
 app = typer.Typer()
 console = Console()
+logger = setup_logger(__name__)
 
 
 @app.command()
-def main():
-    """Console script for keep_gpu."""
-    console.print("Replace this message by putting your code into " "keep_gpu.cli.main")
-    console.print("See Typer documentation at https://typer.tiangolo.com/")
-    run()
+def main(
+    interval: int = typer.Option(
+        300, help="Interval in seconds between GPU usage checks"
+    ),
+    gpu_ids: Optional[str] = typer.Option(
+        None,
+        help="Comma-separated list of GPU IDs to monitor and benchmark on (default: all)",
+    ),
+    vram: str = typer.Option(
+        "1GiB",
+        help="Amount of VRAM to keep occupied (e.g., '500MB', '1GiB', or integer in bytes)",
+    ),
+    threshold: int = typer.Option(
+        -1,
+        help="Max GPU utilization threshold to trigger keeping GPU awake",
+    ),
+):
+    """
+    Keep specified GPUs awake by allocating VRAM and monitoring usage.
+    """
+    # Process GPU IDs
+    if gpu_ids:
+        try:
+            gpu_id_list = [int(i.strip()) for i in gpu_ids.split(",")]
+        except ValueError:
+            console.print(f"[bold red]Error: Invalid characters in --gpu-ids '{gpu_ids}'. Please use comma-separated integers.[/bold red]")
+            raise typer.Exit(code=1)
+        os.environ["CUDA_VISIBLE_DEVICES"] = ",".join(map(str, gpu_id_list))
+        logger.info(f"Using specified GPUs: {gpu_id_list}")
+        gpu_count = len(gpu_id_list)
+    else:
+        gpu_id_list = None
+        gpu_count = torch.cuda.device_count()
+        logger.info("Using all available GPUs")
+
+    # Log settings
+    logger.info(f"GPU count: {gpu_count}")
+    logger.info(f"VRAM to keep occupied: {vram}")
+    logger.info(f"Check interval: {interval} seconds")
+    logger.info(f"Busy threshold: {threshold}%")
+
+    # Create and start Global GPU Controller
+    global_controller = GlobalGPUController(
+        gpu_ids=gpu_id_list,
+        interval=interval,
+        vram_to_keep=vram,
+        busy_threshold=threshold,
+    )
+
+    with global_controller:
+        logger.info("Keeping GPUs awake. Press Ctrl+C to exit.")
+        try:
+            while True:
+                time.sleep(3600)
+        except KeyboardInterrupt:
+            logger.info("Interruption received. Releasing GPUs...")
 
 
 if __name__ == "__main__":
diff --git a/src/keep_gpu/keep_gpu.py b/src/keep_gpu/keep_gpu.py
deleted file mode 100644
index 8d9a0443..00000000
--- a/src/keep_gpu/keep_gpu.py
+++ /dev/null
@@ -1,72 +0,0 @@
-import argparse
-import os
-import subprocess
-import time
-import torch
-
-from .benchmark import run_benchmark
-from keep_gpu.utilities.logger import setup_logger
-
-logger = setup_logger(__name__)
-
-
-def parse_args():
-    parser = argparse.ArgumentParser(
-        description="GPU Idle Monitor and Benchmark Trigger"
-    )
-    parser.add_argument(
-        "--interval",
-        type=int,
-        default=300,
-        help="Interval in seconds between GPU usage checks",
-    )
-    parser.add_argument(
-        "--gpu-ids",
-        type=str,
-        default=None,
-        help="Comma-separated list of GPU IDs to monitor and benchmark on (default: all)",
-    )
-    return parser.parse_args()
-
-
-def check_gpu_usage(gpu_ids=None):
-    result = subprocess.run(
-        ["nvidia-smi", "--query-gpu=utilization.gpu", "--format=csv,noheader,nounits"],
-        capture_output=True,
-        text=True,
-    )
-    usage_lines = result.stdout.strip().split("\n")
-    usages = [int(line.strip()) for line in usage_lines]
-
-    if gpu_ids is not None:
-        usages = [usages[i] for i in gpu_ids if i < len(usages)]
-
-    return any(usage > 0 for usage in usages)
-
-
-def run():
-    args = parse_args()
-
-    if args.gpu_ids:
-        gpu_ids = [int(i.strip()) for i in args.gpu_ids.split(",")]
-        os.environ["CUDA_VISIBLE_DEVICES"] = ",".join(map(str, gpu_ids))
-        logger.info(f"Using specified GPUs: {gpu_ids}")
-        gpu_count = len(gpu_ids)
-    else:
-        gpu_ids = None
-        gpu_count = torch.cuda.device_count()
-        logger.info("Using all available GPUs")
-
-    idle_count = 0
-    logger.info(f"GPU count: {gpu_count}")
-    while True:
-        if not check_gpu_usage(gpu_ids):
-            idle_count += 1
-        else:
-            idle_count = 0
-
-        if idle_count >= 1:
-            run_benchmark(gpu_count)
-            idle_count = 0
-
-        time.sleep(args.interval)